Along with the development of the Internet technology, a great diversity of information exchange manners are available. Due to usability and perfect function, intelligent terminals play an increasingly important role in information exchange. Voice function is an important function in intelligent terminals, such as voice identifying, voice communication, and recording, which are all functions commonly used by users in daily life. However, during propagation, sound loss occurs, and the problems of noise interference and the like exist, which will affect a processing effect implementing the voice tone replay or precise identifying of the abovementioned voice function.
Taking a double-terminal call scenario as an example, for example, in application based on a conference device, a telephones or voice chat, the echo cancellation technology can be adopted, to improve call quality. In the echo cancellation technology, a reference signal and an echo signal are adopted for performing echo cancellation. The so-called reference signal refers to: a far end signal, that is, a voice signal obtained by playing in a speaker playing terminal, for example, a signal generated by decoding a voice sent by a terminal device A by calling a playing thread drive decoding unit. The so-called echo signal refers to: a near end signal, that is, a voice signal collected in a microphone collection terminal, for example, a signal generated by decoding a voice collected at a sound card through calling a collecting thread drive decoding unit by a terminal device B, and then echo cancellation processing is performed in a collection thread. The echo signal can be obtained after the reference signal undergoes attenuation and/or noise interference in the propagation process.
Echo latency exists between the reference signal and the echo signal, in related technology, echo cancellation processing needs to be performed in order to avoid relative latency of the reference signal and the echo signal caused by echo latency. In current technology, echo cancellation processing is usually performed aiming at a current frame of signal, and due to relatively great fluctuation of a difference between the length of a reference signal buffer and the length of an echo signal buffer, an echo path needs to be re-estimated frequently, the stability and precision of an echo cancellation effect are low, and the current frame of signal cannot comprehensively assess the overall sound variation state. Currently, no effectively resolving solution is available for this problem.